Citi: DeepSeek V4 version further focuses on cost and performance trends. The MiniMax-W (00100) is valued at HK$1,330 per share.

Citibank released a research report stating that models such as DeepSeek, MiniMax-W (00100), Kimi, and GLM have expanded the TAM of the industry by reducing experimental and deployment costs; MiniMax-W (00100) is valued at HK$1,330 per share, based on a forecasted 30 times P/S ratio in 2028; and using a two-year expected P/S ratio to reflect the company's strong revenue growth, with a compound annual growth rate of 128% from 2025 to 2030, and 184% from 2025 to 2028. The industry rates the company's risk as a "buy". The report points out that DeepSeek's V4 model, particularly its V4-Pro version, is on par with Claude Sonnet 4.6 in terms of artificial intelligence analysis index, while priced much lower than GPT-5.5, indicating a growing polarization between open-weight models and closed-frontier models. In encoding, intelligent agent workflows, and long-context application cases, the gap between open-weight models has narrowed, primarily due to cost-competitive model architectures. The industry indicated that although DeepSeek's pricing on V4-Pro and V4-Flash is much lower than similar models, the company acknowledges that it still lags behind closed-source models in reasoning ability by 3 to 6 months. However, closed-frontier models can still maintain significant competitive advantages in long-term workflows, as reliable performance is crucial in relevant workflows. With the intensification of open-weight competition, a broader ecosystem will benefit from accelerating enterprise adoption, expanding the total potential market for artificial intelligence (TAM), and raising the threshold for differentiating frontier models.